In recent years, researches on the recognition for Inverse Synthetic Aperture Radar (ISAR) images continue to deepen, while most methods only use the amplitude information of the ISAR image data. Besides, high-order terms in the complex-valued (CV) received signals for maneuvering space targets will cause defocusing on the ISAR images, which affects the accuracy of the recognition. For a steadily rotating maneuvering target, its high-order phase information between frames is relevant, and this information can be used to facilitate recognition. To this end, this letter proposes an end-to-end recognition framework in the CV domain based on the transformer model. It uses multi-scale feature extraction strategy and CV attention mechanism to get the local and global hybrid feature. Besides, A spatio-temporal transformer (STT) block is proposed to obtain the spatio-temporal correlation between image frames to assist recognition. Finally, a residual CNN block is introduced to promote diversity in the captured representations. In the experimental part, the recognition results of the proposed method on the real and simulated dataset are better than those of other methods. Compared with the classic sequence recognition method CVLSTM, the recognition accuracy and kappa coefficient of the proposed method are increased by approximately 5.6% and 5.4% respectively.

Since synthetic aperture radar (SAR) can observe all-weather, it is widely used in ship target detection and segmentation tasks. However, SAR images have complex backgrounds and clutter interference, which affect the segmentation accuracy. This paper proposes a saliency-combined complex-valued U-Net. The network consists of two parts, namely complex-valued U-Net(CV-UNet) and original U-Net. The CV-UNet is used to process the measured data of SAR images which contains amplitude and phase information. The original U-Net is used to process the saliency map generated by the SAR image, and the results of two parts of the network output are connected. The experiment uses the measured data of HISEA-1 to make a target segmentation dataset, and uses the trained network for testing. The results show that the performance of the proposed method is better than that of the original U-Net and CV-UNet.

Synthetic aperture radar (SAR) system is susceptible to electromagnetic jamming during imaging, which seriously affects the subsequent interpretation of SAR images. Aiming at the problem of active suppressive jamming, this paper proposes a suppression method of SAR suppressive jamming based on self-supervised complex-valued deep learning, which consists of a novel complex-valued jamming suppression network CV-MUNet++ and a self-supervised training strategy. CV-MUNet++ could fully use the amplitude and phase information of complex-valued SAR images. The network’s weights, activation functions, and convolution operations are designed for complex domain processing. The different information representations of target and jamming in amplitude and phase in SAR images are mined to achieve jamming suppression. The self-supervised training strategy is proposed to solve the problem of relying heavily on manually labeled samples in the traditional network training process and is suitable for application scenarios where ground truth is difficult to obtain under complex jamming. The experimental results show that the proposed method could effectively suppress the active jamming of complex backgrounds and has the ability to self-supervised intelligent jamming suppression.

In Synthetic Aperture Radar (SAR) images, ship targets suffer from blurring due to pitch, yaw, and sway, resulting in poor recognition accuracy. This paper proposes a complex-valued convolutional neural network (CV-CNN) architecture called CV-RotNet. This neural network could realize the recognition of SAR defocused ship targets without three-dimensional rotation refocusing. CV-RotNet makes full use of the amplitude and phase information of SAR images. Based on the classic deep learning architecture, RotNet and CV-RotNet were designed from the real domain and the complex domain. CV-RotNet and RotNet are tested on the five types of SAR three-dimensional rotating target simulation samples and the three types of GF-3 real ship target samples. Experimental results show that the average accuracy of CV-RotNet is higher than RotNet with the same degree of freedom, which reflects the advantages of CV-RotNet over RotNet.

In synthetic aperture radar (SAR) images, ship targets are defocused due to three-dimensional rotation, which affects subsequent SAR target detection and recognition tasks. This paper proposes a complex-valued convolutional neural network (CV-CNN) structure called CV-RefocusNet to refocus SAR three-dimensional rotating ship targets. CV-RefocusNet includes two parts of feature extraction network and image reconstruction network and adopts an end-to-end design method. To make full use of the amplitude and phase information of complex SAR images, the convolutional layer, deconvolutional layer, and activation function in CV-RefocusNet are all extended to the complex domain. Then refocusing experiments on simulated SAR data and GF-3 SAR data show that CV-RefocusNet could further improve the focus accuracy instead of real-value CNN (RV-CNN) with the same degree of freedom.

Inverse synthetic aperture radar (ISAR) is an effective detection method for targets. However, for the maneuvering targets, the Doppler frequency induced by an arbitrary scatterer on the target is time-varying, which will cause defocus on ISAR images, and bring difficulties for the further recognition process. It is hard for traditional methods to well refocus all positions on the target well. In recent years, generative adversarial networks (GAN) achieves great success in image translation. However, the current refocusing models ignore the information of high-order terms containing in the relationship between real parts and imaginary parts of the data. To this end, an end-to-end refocusing network, named Complex-valued Pix2pixHD (CVPHD) is proposed to learn the mapping from defocus to focus, which utilizes complex-valued (CV) ISAR images as input. A complex-valued instance normalization layer is applied to mine the deep relationship between the complex parts by calculating the covariance of them and accelerate the training. Subsequently, an innovative adaptively weighted loss function is put forward to improve the overall refocusing effect. Finally, the proposed CVPHD is tested with the simulated and real dataset, and both can get well-refocused results. The results of comparative experiments show that the refocusing error can be reduced if extending the pix2pixHD network to the CV domain and the performance of CVPHD surpasses other autofocus methods in refocusing effects. 1The code and dataset have been available online (https://github.com/yhx-hit/CVPHD).

Recently, researches on the classification for inverse synthetic aperture radar (ISAR) images continue to deepen. However, the maneuvering and attitude adjustment of space targets will bring high-order terms to received echoes which cause defocus on ISAR images and affect classification. The current classification models ignore the information of high-order terms containing in the relationship of real parts and imaginary parts of data. To this end, this letter proposes an end-to-end framework, called CV-GNN, specifically for the classification of defocused ISAR images under the few-shot condition. It models the features of real parts and imaginary parts of complex-valued (CV) images as graph information reasoning. Specifically, the deep relationship between them is mined to contribute to classification by complex-valued graph convolution. Moreover, the backpropagation process is derived in detail for updating the weights and bias of the network. The proposed method is then experimented with a mixed few-shot dataset of real and simulated data. Compared with the state-of-the-art methods, CV-GNN performs well in defocused image classification for each class of targets, and ablation studies verify the effectiveness of complex-valued network and graph neural network. The code and dataset will be available online (https://github.com/yhx-hit/cv_gnn).